What is BERT and why is it a big deal?


BERT” stands for Bidirectional Encoder Representations from Transformers. It is a natural language processing (NLP) pre-training system for training systems that answer questions the users ask.

Now, thinking of Google’s search engine as a simple machine is a huge misunderstanding. Anyone opting for black hat SEO techniques in the past decade can tell you that. By now it’s an extremely complex system that is learning from the users on the go. But simultaneously we also pick up habits because of the search engine.

Important: While the system is learning about how we put in our queries, over the years our habits also changed because of our experiences on how well Google can answer our questions.

More specifically, we are learning how to refine the searches: what to add and what to leave out. In most cases including pronouns, conjunctions, definite articles, etc. did little to make your search more likely to provide a more specific result. This is because of how the engine analyzed your search: focusing on the individual words and their order, but not the context.

Meet the new boss, same as the old boss?

Of course we do know that Google has been looking at ways to determine the context of certain words and phrases for years, that is more or less what latent semantic indexing (LSI) keywords are for: basically, keywords that are semantically related to another one. 

If your content includes LSI keywords, Google is more likely to understand what is it about and provide more relevant search results. For example, if an article titled “Apple: which one is the best?” is about smartphones or the fruit.

BERT is a much more sophisticated method for predicting our behavior and interest based on how we search. Once again, the machine got better at understanding language like we do, so it is more likely that if you put in “making friends”, you will be taken to a WikiHow page instead of a behind the scenes video with Jennifer Aniston. You get me.

As Google puts it, transformers are “models that process words in relation to all the other words in a sentence, rather than one-by-one in order”.

So basically what happened is Google just got better at understanding you.

The effects of BERT


The first effects of the algorithm update went largely unnoticed by SEO experts and tools both, but there has been quite a lot of movement since.

The reason most didn’t really pick up on that something big is happening is because it mostly affected long-tail, more conversational queries.

But not only. Here at ROI Foundry, for example, we noticed that one of our articles started to behave quite erratic. To give you context: it is an original research about the most searched online stores and products in Hungary.

What happened is we lost a few dozen keywords that the article previously ranked for – most of which were completely irrelevant, like “pcx webshop”, “alakreform webshop” and so on.

So what we lost were the dozens of keywords entirely irrelevant to the contents of the article if you understand the context. Positions in which the article appeared for searches that had nothing to do with us.

That is arguably positive.

Also, many of our guides started ranking for queries that are extremely relevant to the topic and contents, or if they were ranking before, many jumped ahead dozens of positions.

Yay again!

GOOD TO KNOW: As we see it with our and our client’s sites, BERT didn’t really affect the overall organic traffic anywhere, there are only slight drops.

But based on the movements in ranking keywords and positions, most of that organic traffic just got much more relevant and targeted.

Currently, it seems BERT is settling in. We see pages start to rank for new keywords in the TOP 10 and losing the ranking entirely on the same day. We see articles jumping ahead or backwards 40+ positions suddenly. It will be some time before we know the full extent of how BERT reorganized the SERPs.

Especially since it is rolling out for different languages in a different pace as it learns, as it “takes models that learn from improvements in English (a language where the vast majority of web content exists) and apply them to other languages” as Google puts it.

There is also a thing worth noting with the 10% figure you are seeing.

Note: Google stated that the update affects around 10% of searches, at least in English. This doesn’t mean that 10% of SERPs will be reorganized.

 If you read the original announcement carefully, and I suggest you do, you can also see them noting that 15% of all search queries they process every day are brand new. Overall that 10% figure is just in the ballpark and is more useful for the folks at Google than for us in determining what to expect.

How to optimize for BERT?


Don’t.

In 2015 I had a lecture titled “How to write content that Google likes?”, and my first and second statement were, every single time: “Never do that. Write for humans.”

That is still the best piece of advice I can give you. Poorly-written, short, irrelevant content is being thrown out of the result pages like crazy right now. Content that is truly relevant, useful and high-quality will still be your strongest channel of acquisition.

Focus on the questions your audience is actually asking, and answer them. Use FAQs, how-to guides, Q&As, because the likelihood of those being understood and picked up by the algorithm just increased.

And don’t expect miracles.

As Google reminds you: Even with BERT, we don’t always get it right. If you search for “what state is south of Nebraska,” BERT’s best guess is a community called “South Nebraska.” (If you've got a feeling it's not in Kansas, you're right.)

And as Danny Sullivan is explicitly saying…

 

Just write useful content for your audience…